Blog
-
Why do AI agent attacks keep going for the credential, not the model?
By
Major AI agent attacks keep targeting OAuth tokens, API keys, and standing permissions. Here is the four-step pattern and the architectural fix.
-
Prompt Injection Is an Authorization Problem, Not Just a Model Problem
By
Prompt injection cannot be fully solved at the model layer. Authorization is what limits the damage: scope is the control that still holds when the model is fooled.
-
AI Agent Authorization vs. API Keys: What’s the Difference?
By
An API key is one long-lived, over-broad secret. Agent authorization replaces it with a scoped, time-limited, revocable grant tied to a specific user and agent.
-
What Is Caller-Identity Consent for AI Agents?
By
One API, three kinds of caller: a person, the app’s own AI, and an external agent. Caller-identity consent lets the data owner consent to each one independently.
-
Can an AI Agent Approve Its Own Permission Request?
By
A computer-use or prompt-injected agent can click its own Approve if the consent screen sits where it runs. Why that is not proof of a human decision, and the fix.
-
What Is the Context Access Divide?
By
The Context Access Divide shows why open retrieval matters for AI agents, and why it also needs scoped, user-controlled authorization.
-
What Is Discovery-by-Introspection for AI Agents?
By
Discovery-by-introspection tells an AI agent what it may do at credential-check time: a scope-filtered map of authorized endpoints with field-level schemas, so a read-only agent never even sees write operations.
-
What Is a Self-Describing Credential for AI Agents?
By
A self-describing credential is a token that carries everything an AI agent needs to connect: the agent reads where to exchange it from the token itself, then gets back its scopes, endpoints, and request schemas in one step.
-
MCP Security: How to Let an AI Agent Access User Data Safely
By
MCP standardizes what an agent can do; it does not give end users per-agent, scoped, revocable control. Add user-mediated authorization so the user approves scopes, the credential goes to the user, and every call is checked against the grant.
-
What Is User-Mediated Authorization for AI Agents?
By
User-mediated authorization is a model where the account holder grants an AI agent access directly, and the credential is delivered to the user, not pushed to the agent through an automated channel. Access is scoped, time-limited, and revocable.
Get new posts by email
One short email when a new post goes live. No spam, unsubscribe anytime.